Katja Luck⇑, Sebastian Charbonnier, Gilles Travé⇑
UMR 7242, Institut de Recherche de l’Ecole de Biotechnologie de Strasbourg, Bd Sébastien Brant, BP 10413, 67412 Illkirch, Cedex, France
a r t i c l e i n f o
Edited by Marius Sudol, Gianni Cesareni, Giulio Superti-Furga and Wilhelm Just Keywords:
The canonical binding mode of PDZ domains to target motifs involves a small interface, unlikely to fully account for PDZ-target interaction specificities. Here, we review recent work on sequence con-text, defined as the regions surrounding not only the PDZ domains but also their target motifs. We also address the theoretical problem of defining the core of PDZ domains and the practical issue of designing PDZ constructs. Sequence context is found to introduce structural diversity, to impact the stability and solubility of constructs, and to deeply influence binding affinity and specificity, thereby increasing the difficulty of predicting PDZ-motif interactions. We expect that sequence context will have similar importance for other protein interactions mediated by globular domains binding to short linear motifs.
!2012 Federation of European Biochemical Societies. Published by Elsevier B.V. All rights reserved.
1. Introduction
Interactions between proteins are essential for most of the pro-cesses that happen in living cells. Many protein interactions in cell signalling are mediated by interactions between globular domains and short linear motifs (SLiMs)[1]. SLiMs are short disordered pro-tein sequence segments that often become folded in their bound state [1,2]. Important insights on domain–SLiM interactions have been gained from studies on PDZ (PSD95-DLG1-ZO1) domains.
PDZ domains constitute a large family of globular domains found in prokaryotes and eukaryotes[3]with about 270 occurrences in the human proteome (see next section). PDZ-domain containing proteins are implicated in diverse cellular functions such as estab-lishment and maintenance of cell polarity[4], signal transmission in neurons[5]or in visual and auditive processes in the eye and ear [6,7], cell migration[8], and regulation of cell junctions[9]
(for reviews see[10,11]).
The core PDZ fold adopts an antiparallelbbarrel structure[12]
comprising 5–6bstrands and 1–2ahelices (Fig. 1A). PDZ domains mainly recognize PDZ-binding motifs (PBMs) that are situated at
the very C-terminus of proteins. Some PDZ domains may also bind internal (i.e. non-C-terminal) PBMs [13,14] or lipids [15]. PBMs bind via b augmentation to PDZ domains, e.g. PBMs adopt ab strand that pairs in an antiparallel manner with theb2 strand of the PDZ domain (Fig. 1A). The carboxylate group of the last residue of the SLiM (here, the term peptide will be equally used) is hydro-gen-bonded to backbone amides of residues from the carboxylate binding loop (b1–b2 loop), thereby determining the C-terminal peptide selectivity of PDZ domains (Fig. 1A). Based on the recogni-tion of C-terminal SLiMs, peptide posirecogni-tions are numbered starting from the last residue (position 0, p0) going backwards (p!1, p!2, and so forth). The last residue is almost always a hydrophobic res-idue, mainly Val, Leu or Ile. The third last peptide residue (p!2) can be either Thr or Ser (class I), hydrophobic (class II), or Glu or Asp (class III), thereby defining three main categories of PDZ-binding motifs[16,17]. Thus, recognition of SLiMs by PDZ domains is based on residues of two key peptide positions, p0 and p!2.
Indeed, it has been generally observed that SLiMs have on aver-age less than four defined positions[2]. Given this small binding interface, numerous studies addressed the question about how SLiMs can fulfill the need for specific protein interactions in cell signalling[18–22]. An increasing number of studies now suggests that protein interactions in cell signalling are not only determined by their minimal interacting fragments (e.g. core globular domain 0014-5793/$36.00!2012 Federation of European Biochemical Societies. Published by Elsevier B.V. All rights reserved.
⇑Corresponding authors.
E-mail addresses: katja.luck@unistra.fr (K. Luck), gilles.trave@unistra.fr (G. Travé).
j o u r n a l h o m e p a g e : w w w . F E B S L e t t e r s . o r g
and SLiM) but also by theircontext[23–26]. The context can be either encoded within thesequences of the two interacting pro-teins; or defined by thecellular environment.In our definition, cel-lular context comprises factors that influence the temporal and spacial distribution of proteins, thereby determining when and un-der which conditions (e.g. local concentration) two proteins will meet in order to bind each other[25]. Sequence context comprises the regions in proteins that surround SLiMs or globular domains and that were shown to have an impact on the domain–SLiM inter-actions. Here, we refer to sequence context asextensions, if they oc-cur directly upstream or downstream of the SLiM or domain and if they are not part of other domains. Sequence context that is not considered as extensions consists of neighbouring domains and re-gions that are not in the neighbourhood of the SLiM or domain.
The repertoire of structures on PDZ domains deposited in the Protein Data Bank (PDB)[27]is immense (Fig. 1B) providing us with a unique perspective on the structural diversity that exists within one protein domain family. Given the accumulated knowl-edge about PDZ–peptide interactions, they can serve as a model system to understand how domain–SLiM interactions are influ-enced by their sequence context. In the following, we will review studies that provide insights on extensions of PBMs and PDZs as well as studies that investigated the interplay between PDZ do-mains and their neighbouring dodo-mains.
2. Precision of the total number of PDZ domains in the human proteome
There has been considerable confusion about the total number of PDZ domains in the human proteome (hereafter called human PDZome). Numbers that are frequently referred to in the ‘‘PDZ liter-ature’’ range from about 250 up to 900. Based on the articles that justify the number of human PDZs claimed in their text, we identi-fied three different sources. Articles that claim a total number of 450 human PDZ domains (see suppl. data for references), refer to
to be highly redundant, probably because it has been based on a set of protein sequences containing several isoforms for the same gene that have identical PDZ sequences. By now, SMART decreased the original number from 450 down to 364 (as of February 2012). Spal-ler[29]represents the second source, which claimed a total number of 918 human PDZ domains. In an erratum published four years la-ter, Spaller corrected this number down to 234 and explained that he had been misled by erroneous numbers that were present in the preprint (but not the final version) of Bhattacharyya et al.[30].
Nonetheless, articles are still being published that seem to ignore the erratum[31]. Finally, several independent studies[32–34]that aimed at finding all human PDZ domains for bioinformatic analyses converge on a total number of about 270 PDZs in the human prote-ome without counting alternatively spliced forms of PDZs. This lat-ter number, which we confirmed in our own investigations, is most probably the best estimate of the size of the human PDZome.
3. Extensions of PBMs
In numerous low-scale and large-scale studies as well as bio-computational analyses researchers sought to decipher the speci-ficity rules of PDZ–peptide interactions [19,35–38]. Apart from peptide residues at p0 and p!2, which are the hallmarks for recog-nition by PDZs, the role of other residues of the PBM is more vari-able. Some studies suggested that the contribution of residues at p!1 and p!3 to the overall binding event is minor whereas in other studies PDZ domains exhibited clear preferences for certain residues at these sites over others[39–43]. More and more studies now converge on the idea that PBMs should at least be extended to p!4 as residues at this position have also been observed to signif-icantly contribute to the binding to PDZs[19,35]. By extending the PBM further and further, a few interesting studies indicate that peptide residues up to p!10 are also implicated in PDZ binding.
In the following, we define the core PBM as consisting of the last four residues (these residues clearly bind in the binding pocket of PDZ domains between theb2 strand and thea2 helix) and any longer PBMs will be considered as being extended.
3.1. Residues upstream of the core PBM modulate the binding affinity to PDZ domains
Interesting observations about PDZ–peptide recognition involv-ing an extended PBM were obtained when comparinvolv-ing the bindinvolv-ing of peptides derived from Wnt-signalling proteinbCatenin and in-ward rectifier K(+) channel protein Kir2.3 to the PDZ domain of TIP1 (tax-interacting protein 1). A longbCatenin peptide bound stronger than a short peptide to the TIP1 PDZ[44,45]. The main contributor to this difference in affinity is most probably Trp at peptide position p!5 (in the following, we will write W!5Þ. Muta-tion of W!5to Ala significantly weakened binding, as did mutation of Pro to Ala or Ser in theb2–b3 loop of the PDZ domain facing W!5
[44]. Interestingly, mutation of R!5to Trp in Kir2.3 peptide led to an astonishing increase in binding affinity from 6.4lM to 8.5 nM [46](Table 1). As for the TIP1-bCatenin interaction, W!5in thebPix C-terminus (Rho guanine nucleotide exchange factor 7) contributes to the binding to PDZ1 of SHANK1 (SH3 and multiple ankyrin re-peat domains protein 1) as its mutation to Ala reduced binding affinity, too[47](Table 1).
3.2. Electrostatic interactions between residues of extended peptides and of PDZ domains
The charge of the residue at position p!5 of the C-terminus of
α2
# PDZs Similarity in %
B A
Fig. 1.Available structural information on PDZ domains. (A) Structure of the PDZ domain of AF6 (Afadin) bound to a C-terminal peptide (LFSTEV) derived from Bcr (PDB ID: 2AIN[112]). The secondary structure elements, peptide positions, and the common signature of the carboxylate binding loop (G/G/where/represents a hydrophobic residue) are indicated. Green dashed lines represent hydrogen bonds that are established between Val at p0 and the carboxylate binding loop. Figure was created with Pymol[113]. (B) Diagram that illustrates the number of human PDZ domains for which there is at least a structure of a PDZ domain in the PDB with X % sequence similarity based on local sequence alignments from BLAST searches (e.g.
for 152 human PDZs there is a structure of a PDZ in the PDB with at least 80%
sequence similarity.)
determined the binding of these two proteins to either the PDZ do-main of SNX27 (sorting nexin 27) or the first two PDZ dodo-mains of
position p!5 of GIRK3 with Arg, as observed in IRK1, was sufficient to induce the binding of GIRK3 to PSD95 and to disrupt its interac-Mutagenesis performed to study extended PBMs. The table summarizes the mutational data obtained from studies that analysed interactions between residues from extended PBMs and PDZ domain residues. If available, measured binding affinities are indicated. wt = wild type, Cter = C-terminal extension of PDZ, alt.spl. = alternatively spliced, Nter-PDZ = construct comprising the Nter-PDZ extended at its N-terminus, AA = amino acids, CC = coiled-coil.
gested ionic contacts between theb2–b3 loop of PDZ3 of DLG4 (disks large homolog 4) and of a C-terminal peptide derived from CRIPT (cysteine-rich interactor of PDZ3) that seemed to be impor-tant for peptide binding.
Structural and mutagenesis studies provided evidence for elec-trostatic interactions between PDZ2 (also referred to as PDZ1) of MAGI1 (membrane-associated guanylate kinase inverted 1) and an extended C-terminal peptide derived from Human Papillomavi-rus (HPV) 16 E6[50–52]. Mutation of the negative charges in the b2–b3 loop of PDZ2 to Gln and Asn or Ala significantly reduced the binding affinity as did mutation of the positive charges at posi-tion -4, -5 and -7 to Ala in the E6 peptide[51](Table 1). In another study, the contribution of upstream peptide residues to the binding affinity to PDZ2 of MAGI1 has been more generally assessed. By measuring binding affinity of several peptides of different length, the preference of PDZ2 of MAGI1 for peptides with positive charges upstream the core PBM has been confirmed[26]. This is one clear case where an increase in specificity is driven by interactions be-tween residues of extended PBMs and residues of theb2–b3 loop.
In contrast, theb2–b3 loop of PDZ3 of MAGI1 did not seem to con-tribute at all to peptide binding[26]. The same peptides were also assayed for binding to PDZ3 of the cell polarity protein hScrib revealing that extended peptides generally bound with higher affinity to PDZ3. This study demonstrates that depending on the PDZ in question, interactions between upstream peptide residues and theb2–b3 loop can have different implications on peptide binding[26].
3.3. The sequence of theb2–b3 loop influences peptide binding C-terminal peptides of APC (adenomatous polyposis coli pro-tein), the glutamate receptor subunit NR2B, and HPV18 E6 bind stronger to PDZ2 of DLG1 (disks large homolog 1) as compared to PDZ1 of the same protein [53–55] (Table 1). Several studies suggest that amino acid differences in the b2–b3 loop of these two PDZ domains mainly account for the different binding prefer-ences via interaction with residues of extended PBMs. Mutation of a Gln in theb2–b3 loop of PDZ2 to Pro (the corresponding res-idue in PDZ1) decreased binding affinity for APC[53]. The con-verse mutation (Pro to Gln) directed at the equivalent position of the b2–b3 loop of PDZ1 increased binding affinity of PDZ1 for NR2B whereas mutation of the same Pro of PDZ1 to Ala did not alter the binding affinity[55]. The Pro in theb2–b3 loop of PDZ1 might also be responsible for weaker binding to E6 in com-parison to PDZ2. Liu et al.[54]noticed that the neighbouring res-idues of the Gln in PDZ2 adopt a particular conformation that contributes to peptide binding, and suggested that the Pro in PDZ1 did not allow its neighbouring residues to adopt a similar favorable conformation. One reason for the weaker binding dis-played by PDZ3 to these peptides might be its shorterb2–b3 loop, which does not provide such a platform for extended peptide binding as does PDZ2[53–55].
3.4. Extended PBMs confer dual binding specificity to PDZ domains A few PDZ domains were shown to have dual specificity, e.g.
binding to PBMs of class I and class II. The dual specificity of PICK1 (protein interacting with C kinase 1) has been mainly attributed to specific residues of thea2 helix[56]. In contrast, the dual specific-ity of PDZ3 of the cell polarspecific-ity protein Par3 can be attributed to an extended binding pocket[57]. PDZ3 of Par3 binds to long C-termi-nal peptides derived from both the vascular endothelial Cadherin (class II PBM, 12 residues) and phosphatase PTEN (class I PBM, 11 residues). However, it significantly binds weaker to a shorter
anymore to a shorter (8-residue long) PTEN peptide[58,57](Table 1).
Analysis of the two available NMR structures of these two com-plexes revealed several contacts between negatively charged resi-dues in the extended peptides and positively charged resiresi-dues of theb2 andb3 strand as well as theb2–b3 loop of PDZ3[58,57]
(Fig. 2A and B). Whereas the Cadherin peptide mediated more favourable interactions by means of its last four peptide residues, the PTEN peptide established more favourable contacts to PDZ3 with its upstream residues. At first sight this might be interpreted as an example where a longer binding pocket leads to more pro-miscuous binding behaviour. Yet, class I peptides such as PTEN have to fit very well to the extended binding site with their up-stream residues in order to be bound by PDZ3, a constraint that might only be fulfilled by a few peptide sequences.
3.5. Conformational changes of theb2–b3 loop upon peptide binding As indicated by the previous examples, residues from extended PBMs mainly modulate binding affinity to PDZ domains via inter-action with residues of theb2–b3 loop. Comparison of available apo (unbound) and holo (bound) NMR and crystal PDZ structures revealed that theb2–b3 loop either changes conformation upon peptide binding or remains unchanged. Two examples of the latter case are represented by PDZ2 of DLG1 and Erbin (Erbb2-interacting protein) PDZ of which theb2–b3 loops exist in a stable conforma-tion when no peptide is bound and this conformaconforma-tion remains un-changed upon peptide binding, also when complexed to different peptides. In Erbin PDZ, there is a chain of aliphatic contacts from theb2–b3 loop to theb3 strand and theb4 strand (Fig. 2C). Addi-tionally, N1345 of theb2–b3 loop seems to establish a hydrogen bond to the backbone of the loop. These interactions between res-idues of the Erbin PDZ are probably the driving forces that keep that loop very rigid providing a stable platform for peptide binding.
Theb2–b3 loop of PDZ1 of ZO1 (Zonula occludens protein 1) is equally long as that of Erbin PDZ, and it adopts a similar conforma-tion upon peptide binding, but it displays a different conformaconforma-tion in its unbound form. Here, the loop seems to restructure upon pep-tide binding allowing for accommodation of upstream peppep-tide res-idues (Fig. 2D). Similar observations were obtained for theb2–b3 loop of the TIP1 PDZ domain (Fig. 2E2) and the PDZ of SHANK1 [47].
3.6. Theb2–b3 loop can form an additional peptide binding pocket that accommodates upstream peptide residues
In many of the examples mentioned in the previous subsec-tions, theb2–b3 loop contributed to the formation of an additional peptide binding pocket together with residues from strandsb2,b3, and sometimesb4. In particular, an aromatic residue (mainly F or Y), located right at the beginning of theb3 strand, is often involved in the formation of this additional pocket. This residue contributed to the binding of peptide residues upstream position -3 and seemed to serve as an anchoring point for the structuring of the b2–b3 loop. The conserved aromatic character of this position in the family of PDZ domains suggests that residues at this position might be of more general importance for the structure and func-tion of PDZs[49].
In general, this additional pocket was frequently observed to have hydrophobic character being occupied by upstream peptide residues with large aliphatic side chains such as Trp, Tyr, or Arg (Erbin PDZ, PDZ1 of ZO1, TIP1 PDZ,Fig. 2C–E1, respectively). How-ever, in the case of the PDZ of SNX27, this additional pocket is rather of hydrophilic character being formed by three arginines
Fig. 2.Contacts between residues of extended PBMs and PDZ residues. The table summarizes the contacts that were observed between residues in extensions of PBMs and residues of PDZ domains or their extensions. A structural representation is provided for most of these PDZ–peptide complexes. Colour code: peptide residues in blue, PDZ residues from holo NMR structures in orange, from apo NMR structures in dark-green, from holo crystal structures in yellow, and from apo crystal structures in light-green.
Structural information was used from the following PDB entries: (A) 2K20[58]; (B) 2KOH[57]; (C) 1MFG[61], 1N7T[59], 2H3L[60]; (D) 2H2B, 2H3M[60]; E1: 3DIW[44], 3GJ9[46]; E2: 3DJ1, 3DIW[44], 3GJ9[46], 2KG2[114]; (F) 3L4F[47], 1Q3P[115]; (G) 3QGL[48]; (H) 2KPL[52], 2I04[50]; (I) 3K1R[66]; (J) 2KBS[67]; (K) 3CYY[69]. Figures
E!5of the C-terminal peptide of GIRK3 (Fig. 2G). Another interest-ing example is that of the trimer ofbPix bound to the SHANK1 PDZ.
The homotrimer is formed by the coiled-coil domain ofbPix lead-ing to three closely located C-termini all carrylead-ing the PBM ofbPix [47]. However, due to steric hindrance, only one of the three C-termini is accessible for binding to SHANK1 PDZ [47]. Here, the additional pocket formed by theb2–b3 loop,b2, and b3 strand adopts a hemispheric shape and interacts with one side of the ring of W!5frombPix (Fig. 2F). The other side seems to be covered from solvent by residues from the coiled-coil domains. In that way, W!5
may contribute to the stabilisation of the whole complex.
In most cases, this additional pocket is occupied by peptide residues from position -5. However, it can sometimes be occupied by residues at different positions, such as W!4 for a phage dis-play-derived peptide bound to Erbin PDZ (Fig. 2C)[59], W!6 for a phage display-derived peptide bound to PDZ1 of ZO1 (Fig. 2D) [60], and Y!7 for ErbB2 peptide bound to Erbin PDZ (Fig. 2C) [61]. Together with observations that peptide residues at position -4 can either contact domain residues from theb2 orb3 strand, from a2 helix or from theb2–b3 loop (see table inFig. 2), this demonstrates to which extent peptides can adapt to PDZ do-mains. A rigid definition of pairs of domain and peptide residues as often considered in PDZ–peptide interaction predictors, would not reflect this adaptability and is therefore likely to be an inap-propriate model.
In most cases, this additional pocket is occupied by peptide residues from position -5. However, it can sometimes be occupied by residues at different positions, such as W!4 for a phage dis-play-derived peptide bound to Erbin PDZ (Fig. 2C)[59], W!6 for a phage display-derived peptide bound to PDZ1 of ZO1 (Fig. 2D) [60], and Y!7 for ErbB2 peptide bound to Erbin PDZ (Fig. 2C) [61]. Together with observations that peptide residues at position -4 can either contact domain residues from theb2 orb3 strand, from a2 helix or from theb2–b3 loop (see table inFig. 2), this demonstrates to which extent peptides can adapt to PDZ do-mains. A rigid definition of pairs of domain and peptide residues as often considered in PDZ–peptide interaction predictors, would not reflect this adaptability and is therefore likely to be an inap-propriate model.